Reliabilism without Epistemic Consequentialism

Transcription

1 Reliabilism without Epistemic Consequentialism Abstract This paper argues that reliabilism can plausibly live without epistemic consequentialism, either as part of a non-consequentialist normative theory or as a non-normative account of knowledge on a par with certain accounts of the metaphysics of perception and action. It argues moreover that reliabilism should not be defended as a consequentialist theory. Its most plausible versions are not aptly dubbed consequentialist in any sense that genuinely parallels the dominant sense in ethics. Indeed, there is no strong reason to believe reliabilism was ever seriously intended as a form of epistemic consequentialism. At the heart of its original motivation was a concern about the necessity of non-accidentality for knowledge, a concern quite at home in a non-consequentialist or non-normative setting. Reliabilism s connection to epistemic consequentialism was an accretion of the 80s, and a feature of only one of its formulations in that decade. 1 Introduction Recent literature conveys the impression that epistemic consequentialism has long been a dominant view about justified belief. 1 One source of this impression is the prominence of reliabilism in the post-gettier history of epistemology. While reliabilism has had many critics, it retains many adherents and remains a dominant perspective on the nature of justified belief and knowledge; indeed, it is arguably the leading form of externalism. Accordingly, if reliabilism were a form of epistemic consequentialism, one could reasonably conclude that the latter has been a major force in traditional epistemology since the late 60s and early 70s. But such thinking would be misguided. Reliabilism in its early formulations was not offered as a consequentialist view, or indeed as a normative theory at all. While it took on a relevantly normative formulation in one classic work nearly twenty years after its emergence (viz., Goldman (1986)), this formulation was an accretion, and lacks a claim to dominance. Moreover, reliabilism s apparently consequentialist formulations aren t really analogous to consequentialism in ethics, nor are they best defended in the way consequentialist views are best defended. Such, at any rate, are some main claims to be defended below. More briefly, my goal is as follows. After defending my historical claims, I will argue that whether or not reliabilism was ever seriously intended as consequentialism, it 1 See especially Berker (2013a), Ahlstrom-Vij and Dunn (2015), and Dunn (2015). 1

2 (C1) can naturally be defended without epistemic consequentialism, and (C2) should be defended without epistemic consequentialism, since its most plausible forms are not properly regarded as consequentialist: certain restrictions are needed to solve central problems like the clairvoyance and new evil demon problems, and these lack a fundamentally consequentialist rationale. I will begin with some preliminary clarifications in 2, defend (C1) in 3 4 and (C2) in 5, and take stock and conclude in 6. 2 Consequences and the Structure of Normative Theory Before proceeding to my arguments, I should first explain what kinds of theories I would count as consequentialist, since I will be relying on some assumptions here that are familiar in normative ethics but less familiar in epistemology. Consequentialism is, I take it, a first-order normative theory, not a meta-normative theory. 2 Since there are different kinds of first-order theory, it is worth considering in more detail what kind consequentialism is. Following Kagan (1992, 1997), we should distinguish factoral and foundational normative theories. To understand the point of the first kind of theory, note that many normative statuses (rightness, value, justification...) even within a domain (prudential, moral, epistemic...) are a function of various relevant factors. Factoral theorists seek, in Kagan (1997: 17) s words, to articulat[e] these various normative factors, and discover how they interact to determine statuses like rightness, value and justification. In other words, factoral theories seek to answer the following question of some normative status S in a domain d: (Q1) The Normative Factors Question: What factors bear on whether a target of evaluation in d has S? What, in other words, are the S-relevant factors? Although factoral theories often posit several factors, they can be monistic. One might, for example, claim that the only factor that bears on whether one s act is prudentially right is its contribution to one s pleasure/pain ratio. This claim can be offered factorally rather than foundationally because one could conceivably go further, adopting a more fundamental but still normative explanation of why one s pleasure/pain ratio is the only prudentially relevant factor. For example, one might defend hedonism about prudential rightness on the basis of 2 Though see Copp (2009, 2014) for a different approach. 2

3 (a) agent-relative act consequentialism about prudential rightness, (b) a welfarist view about prudential goodness, and (c) a hedonist account of welfare. At the foundational level, the theorist offering this position is a consequentialist about prudential rightness and a welfarist about goodness-for. This theorist also believes that welfare is grounded in pleasure, and hence that pleasure is the only factor relevant to prudential rightness. Despite not being meta-normative positions, foundational normative theories do more than answer (Q1). They seek to answer: (Q2) The Normative Foundations Question: What is the fundamental normative explanation of why the S-relevant factors are S-relevant factors? Foundational normative ethics hence aims to provide the most fundamental normative explanations of why various normatively relevant factors are the relevant ones. But not all argumentation in normative ethics occurs at this level. Much just concerns which factors are morally relevant. Consider partiality: can the fact that I am in some relationship with X affect how weighty a moral reason I have to benefit X? It is unclear that answering this question requires having already taken a stand on the correct foundational normative theory. 3 Consequentialism in its most controversial and significant form is, I take it, a foundational normative theory. It must be understood in this way to be inconsistent with paradigmatic nonconsequentialist theories. For as I will emphasize again later, the best interpretations of Kantian ethics arguably make it coincide in its predictions with certain forms of consequentialism, but it still differs in deriving these predictions from a more fundamental non-consequentialist norm; for example, Wood (1999), Herman (1993: Ch.3; 2007: Ch.11), and Korsgaard (1996: Ch.4) suggest that one can derive the duty to promote the well-being of other people, or to accept rules that would prevent suffering, from more fundamental duties of respect or love for persons. Indeed, as Dreier (1993) pointed out, arguably any plausible normative theory is extensionally equivalent to some version of consequentialism, but this hardly establishes the 3 Methodological generalists would say that we must do the foundational work first. But one needn t be a Dancy-style particularist to prefer methodological particularism, insisting that we should give a foundational story only after doing the factoral work. 3

4 truth of consequentialism. For it remains an open question whether the explanations given by the non-consequentialist are preferable. Hence the issue of explanatory priority emerges as the deeper disagreement between consequentialists and non-consequentialists. 4 This fact will prove important in my argument in 5. In its central versions in ethics, consequentialism seeks a certain sort of normative explanation of facts about rightness and related normative properties in terms of value: in particular, an instrumental explanation that reveals a means-end link between rightness and the promotion of value. 5 As Berker (2013a: 343) notes, this qualification on the form of the consequentialist s value-based explanation is important. There are theories that are value-based but not consequentialist, like virtue ethics and Herman (1993) and Wood (1999) s interpretations of Kantian ethics, which suggest that the value of humanity or the value of the good will is normatively fundamental, but which insist that the fundamentally proper response to this value is to respect it. In its direct form, consequentialism about right action upholds the following claim: Direct Consequentialism about Right Action: The fundamental normative explanation of why right actions are right is that they are conducive to final value. In its broadest indirect form, consequentialism about right action upholds the following claim: Indirect Consequentialism about Right Action: The fundamental normative explanation of why right actions are right is that they are produced by something that is conducive to final value, where this something might be the general acceptance of a certain rule (rule consequentialism) or the operation of a certain motive (motive consequentialism). Now, it is important to add two qualifications to these preliminary glosses. Firstly, it is worth stressing that consequentialists of both sorts have for crucial reasons understood value-conduciveness in terms of both the proximal and distal consequences of an act, rule, or motive. Of course, to solve the problem of demandingness, a switch from actual to expected consequentialism might be needed in theorizing about decision procedures rather than right-making characteristics, 6 4 See esp. Scanlon (1998: Ch.2) for this point. As Scanlon argues, consequentialism rests on a misunderstanding of what intrinsic value fundamentally demands, by treating it as fundamentally to be produced despite the fact that many plausible examples of intrinsic values call fundamentally for respect or appreciation (e.g., personhood, friendship, art), and only derivatively for production. 5 Here means is used broadly to encompass not just efficient causal means, but also constitutive means (e.g., pleasure might be a means to well-being in virtue of constituting well-being, on hedonist views). 6 See Bales (1971) and Railton (1984) for this distinction. 4

5 where the latter are the real focus of foundational theorizing. But making this switch doesn t involve claiming that only the expected proximal consequences matter. One can easily have expectations about some distal consequences. And such a theory would anyway be strange: if what fundamentally matters is bringing about the good, why would one ignore the potentially massive outweighing of proximal goodness by distal badness? Not even a desire to appease the time-biases of ordinary agents would justify such a restriction, since consequences can be distal while still being short-term and expected. No direct consequentialist of either a maximizing or a satisficing sort would claim that moving one s finger is right when it would predictably lead to pointless near-term though not immediate disaster. For example, the fact that the only proximal consequence was the pressing of the pretty red button (which the agent enjoyed) is no justification if the distal consequences involve nuclear war. Similarly, no indirect consequentialist of either a maximizing or satisficing sort would claim that it could be right to act in accordance with a rule whose general acceptance would predictably lead to pointless near-term though not immediate disaster. Nor will agent-relative versions of consequentialism embrace these implications except for agents with eccentric utility functions. Such agents will still take distal consequences into account: they will just give them less weight. Other kinds of discrimination may appear in factoral theorizing or theorizing about decision procedures, but discrimination always invites explanation. 7 From a foundational consequentialist perspective, bad consequences of the relevant object of evaluation cannot be ignored until they are outweighed by good ones. Littlejohn (2018: 25) offers a helpful way of capturing this point without excluding certain paradigmatic consequentialist theories. He suggests that consequentialists are committed to the totalizing assumption that the total consequences of whatever is directly evaluated by the theory must figure in the determination of rightness. It is left open to consequentialists to decide how to assign (dis)value to these many consequences. But the only way to practically exclude distal consequences would be by assigning neutral value to them. As Littlejohn emphasizes following Foot (1985) this totalizing assumption is essential for capturing the intuitive appeal of consequentialism: the reason that consequentialism seems so compelling is 7 Indeed, Sidgwick (1907: ) the most consistent of the classical utilitarians took this point to reflect a self-evident principle of rationality. 5

6 that when it comes to the good, the better, and the best, it is irrational to prefer some acknowledged lesser good to one that is greater (25). If one agrees that X s distal badness would be much worse than its proximal goodness, it would seem one prefers the acknowledged lesser good in preferring X (unless there are no alternatives with better distal consequences). Again, one could imagine an eccentric agent-relative view that was indifferent to distally produced suffering while assigning disvalue to proximally produced suffering. An agent who did what this view required wouldn t be irrational in the narrowest sense (i.e., akratic). But the view would achieve this prediction not by claiming that distal consequences are no part of the calculation, but rather by taking them into account and assigning them neutral value. This last point is important. Alhstrom-Vij and Dunn (2018: 1) suggest that ethical egoism is not a totalizing theory, because it only pays attention to the good that is brought about for the agent. 8 But ethical egoism does pay attention to the total consequences. It just evaluates those consequences solely in terms of their goodness for the agent. It is not that some consequences are ignored: instead, only certain values are brought to bear in the overall evaluation of the total consequences. So, contrary to what they suggest, there is no familiar version of consequentialism in ethics that rejects the totalizing requirement. Such a theory would abandon what is arguably the most attractive feature of consequentialism, and the one needed to explain its appeal. So I think it would be better, as Ahlstrom-Vij and Dunn temporarily acknowledge in summarizing Wedgwood (2018) s view, to call a view that rejects the totalizing requirement teleological but not consequentialist. 9 My second qualification stems from a further point that Littlejohn (2018) emphasizes. 10 As he observes, following Pettit (1991: ; 1997: ) and Scanlon (1998: 79-80), there is a deeper reason why consequentialists take norms to be explained by means of conducivity to 8 They seem to suggest this when they contrast act utilitarianism with ethical egoism: Notice several other commitments of act utilitarianism. First, it maintains that rightness is determined not just by some of the consequences of actions but by the total set of consequences... Not all views that have been said to be instances of consequentialism share these features. We can, for example, imagine a form of ethical egoism, where actions are right when maximizing the good for the agent herself (1). 9 See Alhstrom-Vij and Dunn (2018: 3). Wedgwood (2018: 90) is clear on the importance of the totalizing requirement for consequentialists: [C]onsequentialism is best understood as the doctrine that the value or normative status of the relevant states of affairs (whether these states of affairs be acts, or beliefs, or anything else) is derivative from the value of these states of affairs total consequences. 10 This point is also briefly noted by Berker (2013a: 343) but not sufficiently considered in his case for the pervasiveness of epistemic consequentialism. 6

7 fundamental value. The reason is that they understand fundamental value in a certain way: what it is to be fundamentally valuable is to be something that is to be promoted for its own sake (where promote means bring about, not advertise or campaign for). Value-based theories can reject this assumption, but they thereby cease to be consequentialist. With these two qualifications in mind, I think it is most faithful to the ethics literature to restate direct and indirect ethical consequentialism in the following ways: Direct Consequentialism about Right Action: The fundamental normative explanation of why right actions are right is that their total consequences sufficiently promote fundamental value, where fundamental value just is value that is to be promoted for its own sake. 11 Indirect Consequentialism about Right Action: The fundamental normative explanation of why right actions are right is that they are produced by something whose total consequences sufficiently promote fundamental value, where this something might be the general acceptance of a rule (rule consequentialism) or the operation of a motive (motive consequentialism). We can now see what it would take for a theory of justified belief to be consequentialist in a way that yields a faithful analogy with ethics. By analogy with the foregoing formulations of consequentialism about right action, it could take a direct or indirect form: Direct Consequentialism about Justified Belief: The fundamental normative explanation of why justified beliefs are justified is that their total consequences sufficiently promote fundamental epistemic value, where fundamental epistemic value just is value that ought epistemically to be promoted as an end. Indirect Consequentialism about Justified Belief: The fundamental normative explanation of why justified beliefs are justified is that they are produced by something whose total consequences sufficiently promote fundamental epistemic value, where this something might be the acceptance of a certain epistemic rule or the operation of a cognitive process or disposition. To the extent that consequentialism has plausibly had any significant history in epistemology, it is in the second of these forms. But as we will see, there are some striking disanalogies between indirect consequentialist theories in ethics and the epistemological theories that some 11 Here and below, sufficiently isn t meant to exclude maximizing consequentialism, which could be understood as imposing a perfectionist bar on being good enough. 7

8 have claimed to be indirectly consequentialist. In what follows, I will assume that if the analogy is strained too far so that views resembling paradigmatic non-consequentialist theories end up being called consequentialist it will not be worthwhile to regard them as genuinely consequentialist. For example, I will assume that a theory is at best dubiously worthy of the name consequentialist if it embraces at the fundamental level restrictions or side-constraints on promoting the good, or if it insists that value is fundamentally to be respected, not promoted. Because of its role in sustaining the appeal of consequentialism, I will also assume that the totalizing assumption is essential to consequentialism, where adjustments can be made to the underlying value theory to mimic non-totalizing theories. These assumptions do not involve unfair deck-stacking. If epistemologists claiming the label epistemic consequentialism reject the analogues of claims that are essential for motivating consequentialism in contrast to the various non-consequentialist theories, their use of the label threatens to become misleading at best, and merely homonymous at worst. The characterization I ve given above captures what other central players in normative ethics agree to be key features of the view when understood as a foundational normative theory (which is the only format in which it must disagree with Kantianism or virtue ethics). This characterization is also embraced by Berker (2013a), though I believe he is too fast in judging theories to conform to it. As I will discuss in considering Goldman s reliabilism, my characterization does conflict in one respect with the ultimate characterization given by Alhstrom-Vij and Dunn (2014, forthcoming), though they begin with a characterization like my first. But their characterization jeopardizes the analogy with ethics. Moreover, the reasoning that first led them to this characterization in their (2014) involved assuming that reliabilism is a form of epistemic consequentialism. While they treat this assumption as non-trivial in Alhstrom-Vij and Dunn (forthcoming), their willingness to abandon paradigmatic features of the consequentialism/non-consequentialism contrast in ethics is not defensible. Now, when consequentialism is understood as a foundational normative theory, it is possible for some views to appear consequentialist from a certain angle without really being consequentialist. Note that factoral theorizing will plausibly underdetermine foundational theorizing even if our factoral inventory is highly constrained. This underdetermination probably won t leave 8

9 the foundational options wide open. But to expect no underdetermination between at least one consequentialist theory and at least one non-consequentialist theory seems optimistic. If life turns out even a little hard, we will find ourselves choosing between a consequentialist theory and a non-consequentialist theory that from a distance may sound consequentialist because it agrees at the factoral level with some consequentialisms. This fact plays a central role below. The evidence for some forms of reliabilism may look like evidence for epistemic consequentialism, since it is also predicted by epistemic consequentialism. But the best foundational explanation of this evidence and the best underpinning for reliabilism is non-consequentialist. There is a final assumption I should make explicit, which should be obvious but sometimes seems neglected in epistemology. The assumption is that consequentialism is a normative theory, in a broad sense of normative that includes both theories of rational decision procedures and theories of right-making characteristics. Note that there are all manner of theories in philosophy that ground certain kinds of facts about an X in facts about X s effects, or in facts about the effects of something else. Since effects are consequences, these theories ground certain kinds of facts in facts about consequences. Many of these theories are obviously not in the same ballpark as consequentialism. Consider, for example: Causal theories of property identity e.g., the view that F and G are identical iff F and G confer the same causal powers on all particulars that instantiate them. Causal theories of perception e.g., the (admittedly oversimplified) view that a visual experience is a seeing of x iff x causes that visual experience. Causal theories of intentional action e.g., the (admittedly oversimplified) view that a bodily movement is an intentional action iff it is caused by an intention. Response-dependent theories of color e.g., the view that X is red iff X causes redexperiences in observers in normal conditions. It might be heuristically useful to analogize these theories with forms of consequentialism. But it would be a mistake to frame this analogy as the discovery that consequentialism is a popular view in the metaphysics of color, property identity, perception, and action. Such framing would be confused partly because color, perception, and action seem non-normative. As a result, these causal theories aren t in the same business as consequentialist theories. But even if there were some sense in which color and action, say, were normative, it still wouldn t follow that the 9

10 explanations given by causal theories of color and action would be normative explanations. For the relation between perception and its causal constitution as well as between action and its causal constitution is not normative, while the relation between an act s rightness and what makes it right in first-order terms is normative. These points matter in assessing whether reliabilism is rightly viewed as a form of consequentialism. Reliabilism in its earliest formulations was a theory about the nature of knowledge. Knowledge might just be a generic factive mental state. Even if this state were in some sense normative, a causal theory of its constitution analogous to the causal theory of perception would not be a consequentialist theory. For the constitution relation in play needn t be a normative relation of the kind in play in first-order ethics. 3 Reliabilism s Life without Consequentialism (and Occasional Dalliance with It) I turn now to a historical argument for (C1). Here in outline are the main points this section will make. Reliabilism lived for more than ten years as a non-normative theory. It was treated in Goldman (1979) as similar in aim to various first-order ethical theories. But Goldman (1979) s theory could be endorsed factorally or foundationally, and if factorally endorsed it could be embedded within a non-consequentialist foundational view e.g., virtue epistemology, which was supposed to be analogous to one main alternative to consequentialism in ethics. 12 While Goldman (1986) later proposed an analogy between his project and the rule consequentialist s, this was just one moment in the history of reliabilism, and much of Goldman s later work didn t build on the analogy. Anyway, as I will suggest, Goldman (1979, 1986) s views aren t analogous to indirect consequentialism: the analogy is flawed in ways that cannot be ignored if we are serious about our comparisons with ethics. Finally, other reliabilists have continued to hold a more traditional, non-normative form of reliabilism. So although Goldman s contributions are preeminent, one shouldn t forget that reliabilism receives notable treatment in other hands. 12 Goldman (1993: 274) suggests this framing. 10

11 I turn to document these points in more detail. 3.1 Pre-1979 Reliabilism: Non-Accidentality and the Constitution of Knowledge Reliabilism started out as a theory of knowledge, not a theory of justification. 13 What the early versions of reliabilism in Armstrong (1968, 1971), Unger (1968), Dretske (1971), and Goldman (1975) have in common is an attempt to capture the non-accidentality condition on knowledge. While it is standard after Pritchard (2012) to distinguish anti-luck conditions from ability conditions, and reliability conditions are arguably best understood as versions of the latter, this distinction is new. 14 Instead of taking it to show that non-accidentality approaches are not necessarily reliabilist, we could take it as an invitation to distinguish different types of reliabilist theory say, process reliabilism on the one hand and counterfactual reliabilism, to use Vogel (2000) s term, on the other. Indeed, there is plausibly a sense of accident for which it is necessarily true that it s no accident when reliable thinkers beliefs turn out true. This fact remains even if environmental luck cases show there to be some other condition on knowledge, negatively connected to another sense of accident. 15 If one is a reliabilist of the classic stripe, it would be coherent for one to either withhold belief in epistemic consequentialism or reject it. The epistemic consequentialist would derive her reliability condition on knowledge from three things: the assumption that knowledge implies justified belief, an indirect epistemic consequentialist account of justified belief, and a veritist epistemic axiology. The traditional reliabilist needn t accept the reliability condition for such reasons. She can be a reliabilist simply because she thinks knowledge is non-accidentally accurate belief, and reliability provides the best unpacking of such non-accidentality. There is no necessary tie between embracing a view that sees non-accidentality and the 13 Ramsey (1931b: ) provides an interesting exception in giving a reliabilist account of reasonable belief, and there is evidence that he was an epistemic consequentialist (see esp. p.196). But as Goldman and Beddor (2015) observe, Ramsey s early reliabilism attracted no attention at the time and apparently did not influence reliability theories of the 1960s, 70s, or 80s. 14 Goldman and Beddor (2015) describe Goldman (1976) s response to the fake barns case as counterfactual reliabilism, however. 15 But see Millar (2010) and Littlejohn (2014) for a derivation of the anti-luck condition from the ability condition. 11

12 main or even the only thing one must add to true belief to get knowledge and embracing epistemic consequentialism. Indeed, a close parallel in ethics to early reliabilism s emphasis on non-accidentality is provided by interpretations of Kant that see non-accidental conformity to moral law as the defining feature of moral worth, and what is missing in examples like the shopkeeper case, which Kant (1785/2012: 4.398) describes as a case of fortunate (i.e., accidental) rightness. 16 Indeed, the best recent work explaining the thought that moral worth is a matter of doing the right thing for the right reasons has itself been inspired by the AAA pattern from virtue epistemology: according to Mantel (2018) and Lord (2018), for example, an act only has full moral worth if it is right as a manifestation of reasons-sensitive dispositions, which both suggest is a kind of aptness in Sosa s sense. It is not enough, according to these theorists, to say that moral worth consists in the coinciding of right-making reasons with one s motivating reasons (as Markovits (2010) suggested): the coinciding itself must manifest reasons-sensitive dispositions. One can also be a reliabilist of the classic stripe without offering reliabilism as a normative theory of any kind. Armstrong (1968: Ch.9) s first presentation of his reliabilism about knowledge appeared in a book on the metaphysics of mind, where he treated knowledge a mental state and saw perceiving to be a special case of knowing (see (1968: Ch.10) and cf. (1961)). On this kind of view, one could hold that knowledge is no more normative than the factive mental states that are its determinates. 17 If so, the condition of non-accidental arrival at the truth will just be a generalization of the condition of non-accidental veridicality on seeing that p. It no less a mistake to claim that this kind of view is an example of consequentialism in epistemology than it is to say that the causal theory of properties is a version of consequentialism in metaphysics. It is worth adding that even if knowledge were clearly a normative relation in the same sense in which paradigmatically properties like justification and obligation are normative, it still wouldn t follow that a metaphysical account of the nature of knowledge is a normative theory in the sense in which consequentialism is a normative theory. Ethical theorists commonly hold that giving an account of the constitution of something normative is different from giving a 16 For some different versions of this interpretation, see Herman (1981), Benson (1987), Baron (1995), and Stratton-Lake (2000). 17 For further defense of this view, see Sylvan (2018). 12

13 first-order normative theory. First-order normative theories do not give analyses of normative properties a task reserved for meta-ethics but rather substantive theories that explain in a distinctive normative way why certain acts or attitudes have these properties Reliabilism It must be acknowledged that reliabilism did in the hands of Goldman (1979) become a normative theory in a broad sense. Goldman was explicit from the first page: The term justified, I presume, is an evaluative term, a term of appraisal. Any correct definition or synonym of it would also feature evaluative terms. I assume that such definitions or synonyms might be given, but I am not interested in them. I want a set of substantive conditions that specify when a belief is justified. This might be defined in other ethical terms or phrases, a task appropriate to meta-ethics. The task of normative ethics, by contrast, is to state substantive conditions for the rightness of actions. Normative ethics tries to specify non-ethical conditions that determine when an action is right. [... ] Analogously, I want a theory of justified belief to specify in non-epistemic terms when a belief is justified. 18 On the same page, Goldman mentions act utilitarianism as an example of a theory that gives substantive conditions for rightness. But he doesn t suggest that his view is analogous to utilitarianism in more than the general sense clarified in the above passage. We are not getting the overt modelling on consequentialism here that we get in Goldman (1986). Still, one might think that the normative theory Goldman (1979) s view best resembles is a monistic indirect consequentialism say, rule utilitarianism or, closer yet, motive utilitarianism. For this reason, one might not find it misleading to say that indirect epistemic consequentialism received its first detailed formulation in Goldman (1979). But even this claim lacks plausibility. Goldman (1979) s theory is in key ways not parallel to indirect consequentialism. There are at least three features of Goldman (1979) s account that support this verdict, on which I ll comment briefly in outline and then at greater length: A. The account counts only certain highly restricted outputs of processes in the measure of their reliability, which is not what one would expect in a theory analogous to any familiar consequentialism in ethics Goldman (1979: 1). 19 Goldman (1979: 12 13). 13

14 B. Goldman s account of defeat has no analogue in any familiar consequentialism in ethics. While it is a more purely reliabilist account of defeat than the one found in some of his later work (see, e.g., Goldman (1986) and Goldman (2011)), it is not thereby more consequentialist, but less so. C. The asymmetry the account yields in the conditions for justified basic belief vs. justified non-basic belief has no real analogue in consequentialism in ethics. 20 The closest analogue of the restrictions Goldman places on non-basic beliefs are side-constraints, which are paradigmatically non-consequentialist. Let me now elaborate on each, taking (A) first. While the reliability of a process is determined by looking at certain properties of its consequences, only a highly restricted class of consequences matters viz., only the immediate outputs of the process, which Goldman (1979: 13 14) takes into account in his clauses (6A) and (6B). This restriction lacks precedent in the consequentialist tradition in ethics. It conflicts with the totalizing requirement mentioned earlier as essential for explaining the intuitive appeal of consequentialism. Direct consequentialists take all consequences of an act to be relevant, and indirect consequentialists take all consequences of whatever they directly evaluate in consequentialist terms rules or motives into account. Moreover, while one could imagine a factoral theory that restricted consequences to immediate ones, it is hard to tolerate such a restriction as a piece of fundamental normative theorizing that is genuinely consequentialist in spirit. Why should the proximal/distal distinction have such significance? The nonconsequentialist, of course, has an answer that would derive this restriction from some more fundamentally appealing restriction. The consequentialist does not, unless distal consequences are considered but given less weight in the associated value theory. Now, Ahlstrom-Vij and Dunn (2014) try to use the presence of the immediacy restriction in reliabilism to answer Berker (2013a) s argument against epistemic consequentialism, insisting that reliabilism is both a direct and an indirect consequentialist theory. But they begged the question there by assuming without argument that reliabilism is a form of epistemic consequentialism, 21 and by using direct in a sense not used in the consequentialist literature in ethics. Because they assumed there that reliabilism is a form of consequentialism, they assumed that the immediacy restriction is available to the consequentialist in virtue of its availability to 20 Goldman (1979: 13 14). 21 To be fair, Berker (2013) also paints reliabilists as epistemic consequentialists. 14

15 the reliabilist. But it is not available to reliabilists for consequentialist reasons, if one uses consequentialist in a sense that parallels ethics. While Ahlstrom-Vij and Dunn (forthcoming) now acknowledge that it is an open question whether reliabilism is a kind of consequentialism, they end up admitting that Goldman s view contains the epistemic analogue of side-constraints. They also do not give a precedent in the consequentialist literature for the immediacy restriction: their example of ethical egoism is problematic for reasons explained earlier. A third feature of Goldman s account that is hard to derive from consequentialist foundations is its theory of defeat. Of course, the very idea of defeat is not foreign to consequentialism: an action can be right in one circumstance but wrong in another if it has different consequences in the two. Examined from a distance, adding or subtracting consequences might resemble adding or subtracting information. But the particular way in which reliabilists model defeat is not what one would expect if reliabilism were given consequentialist foundations. Goldman (1979) s account of defeat is especially interesting to consider, since it is his only purely reliabilist account of defeat. He moved in (1986) to allow beliefs and experiences to be defeaters irrespective of ancestry, and continued in (2011) to give experiences this role. Goldman (1979) s account of defeat is that a belief produced by a reliable process has prima facie but not ultima facie justification if there is an alternative reliable process available to the believer that would have resulted in a different doxastic attitude (disbelief or suspension). There is no direct or indirect consequentialist analogue of this account of defeat. There is no direct consequentialist analogue of this conception of defeat, since direct consequentialism is forward-looking and the general value of the propensities of act-producing processes are swamped in rightness evaluation by the particular consequences of the token act. Perhaps there is a vague analogy between the relevance of alternative processes and the relevance of alternative strategies open to the agent at the time of action. But note that if two strategies that recommend different acts would have equally good consequences, act consequentialism permits either; by contrast, if one reliable process would recommend belief and an equally reliable process would recommend suspension of judgment, process reliabilism doesn t permit either belief (and rightly so) Cf. Littlejohn (2012: 77), who notes that epistemic consequentialism fails to secure this prediction but maligns Goldman because he assumes Goldman is a consequentialist. 15

16 For a related reason, there is also no indirect consequentialist analogue of Goldman (1979) s conception of defeat. Consider rule consequentialism first. If two rules that recommend different acts A and A* in some circumstance C are such that their acceptance would have equally good consequences, or if two rules are equally reliable guides to promoting the good, it is permissible to do either A or A* in C. Something similar goes for motive consequentialism: if two motives are equally reliable guides to promoting the good, or are such that acting on either would tend to produce similar amounts of goodness, then acting on either is permissible. But again, if one reliable process would recommend belief and an equally reliable process would recommend suspension of judgment, process reliabilism (rightly) doesn t permit both. Curiously, the less purely reliabilist stories Goldman later tells about defeat are more similar to what consequentialist foundations would encourage. Note that the consequentialist can help herself to the structural distinction between subjective and objective rightness. She might hold that justified action requires subjective rightness, and hence must have good expected consequences, or be in accordance with a rule whose acceptance would have good expected consequences. Goldman s later treatment of beliefs and experiences as potential defeaters might be read in a similar way: if some beliefs or experiences can influence expected reliability, they can thereby impact epistemic justification. But these proposals about defeat in epistemology are often regarded as ad hoc: as a thoroughgoing externalist view, reliabilism only explains why actual reliability matters, not perceived or apparent reliability. 23 A final feature of Goldman (1979) s account that doesn t straightforwardly follow from consequentialist foundations is its (modest) foundationalist structure, and specifically the asymmetry it yields between the justification conditions for basic and non-basic beliefs. Of particular interest is the requirement in Goldman s recursive account that non-basic beliefs not be based on unjustified beliefs on pain of inheriting this unjustifiedness. The analogue of this requirement in the practical case would be a requirement that non-basic actions not be based on wrongful actions. Now consider the practical basing relation of immediate interest to the consequentialist, which is the means-end relation. If consequentialism were strongly parallel to Goldman s reliabilism, we would expect the consequentialist to claim that a non-basic action 23 See, e.g., Greco (2010: 158) 16

17 cannot be right if it is performed via wrongful means. But consequentialists shouldn t like this claim. While a consequentialist can allow that a killing of an innocent person is pro tanto bad and prima facie wrong in virtue of realizing an intrinsically bad state of affairs, a killing can become right ultima facie if its effects are good enough. While some consequentialists set high thresholds, they will eventually allow tradeoffs. Otherwise they admit side-constraints, and as I will explain later, embracing side-constraints in a foundational theory makes one a non-consequentialist. The thinking here is the reverse of what seems sensible in the epistemic case. The all-things-considered deontic evaluation of the basis belief is prior to the evaluation of the based belief, while the all-things-considered deontic evaluation of the means action is posterior to the evaluation of the goal action. Note again that I am not saying that there won t be heuristically interesting analogies between the foundationalist s structure for justified beliefs and some normative structures that consequentialists don t frown upon. There is, for example, an interesting analogy between regress arguments for foundationalism and regress arguments for intrinsic value. But what we need is not some indirect analogy between a theory of epistemic justification and a theory of value, but rather an analogy between a theory of epistemic justification and a theory of right action. For there to be an analogy strong enough to say that the same view appears in both domains, we need more than rough resemblances between unlike normative categories. What we need are parallels between like normative categories i.e., justified belief and justified action. Goldman did draw an analogy between reliabilism and rule consquentialism in Epistemology and Cognition. But despite this fact, many of the disanalogies to which I ve drawn attention remain: Goldman (1986) keeps similar restrictions on the kinds of processes and outputs that matter, and preserves an asymmetry between basic and non-basic beliefs like that in Goldman (1979). Given those remaining disanalogies, Goldman s view is more similar in the end to moderate deontology While that claim might sound strange, don t forget that deontologists agree that there is a strong reason to promote the good: they just think this reason can be defeated if the only way to promote the good would be by violating restrictions. 17

18 3.3 Non-Goldmanian 80s Reliabilism Epistemology and Cognition was not the only memorable event for reliabilism in the 80s. A different reliabilism indicator reliabilism also flourished in that decade. While the indicator reliabilism of the time was overshadowed by Goldman s approach, insights from 80s indicator reliabilist theories appear in Goldman s recent views (e.g., Goldman (2011)). As I ll argue shortly, if indicator reliabilist theories are counted as consequentialist, then views in ethics that aren t consequentialist will be wrongly classified as consequentialist. Besides indicator reliabilism, the 80s also featured reliabilist theories of knowledge that like Goldman (1967) s causal theory denied that justification is necessary for knowledge and are consistent with thinking of knowledge as no more normative than factive mental states like seeing that p. 25 So while Epistemology and Cognition was a classic moment in the history of reliabilism, it would be wrong to think that reliabilism took a definite turn toward consequentialism in the 80s. Let s further consider indicator reliabilism, and then briefly consider the clearer case of 80s reliabilism about knowledge. Foreshadowed by some externalist accounts of knowledge in the 70s, 26 indicator reliabilism about justification received subtly different formulations in Swain (1981a, 1981b, 1985) and Alston (1985, 1988). Alston s account is more intuitive, so I ll take it as the paradigm. Alston s account is a two-level account that first identifies justified beliefs with beliefs based on adequate reasons, and then identifies adequate reasons with reliable indicators. I do not think that the structure of this theory automatically recommends a consequentialist interpretation. If incorporating reliability in an account of justification in this way were sufficient for making the account consequentialist, then ethical theories that are not consequentialist will be wrongly classified as consequentialist. I will take an appealing combination of views in ethics to illustrate this point. I find it attractive to combine (a) an account of practical justification in terms of possessed objective reasons 25 I have especially in mind Dretske (1981). While Foley (1987) famously defended an internalist theory of rational belief, he also recommended divorcing the theory of knowledge and the theory of justified belief and indicated his sympathies for a reliabilist theory of knowledge. 26 See Dretske (1971) and Armstrong (1973) s thermometer view, a name unwittingly hearkening back to an indicator reliabilist theory of meaning by the same name that Price (1953: ) attacked. Sellars (1956: VII.31) approvingly alludes to Price s attack just before anticipating and attacking the indicator reliabilist account of knowledge in VIII. 18

19 with (b) an account of objective reasons for action that analyses them in terms of objective evidence and rightness a la Thomson (2008) and Star (2015). This account is structurally like Alston s: (a) is parallel to the first layer of his view, and (b) is parallel to the second layer, with objective evidence being naturally construed as the same thing as a reliable indicator. A deontologist can hold (a) and (b). Thomson is a paradigmatic deontologist, and she accepts (b) and could accept (a). Star contrasts his view with indirect consequentialism but embraces (a) and (b). Consistency is possible here for many reasons, one of which is that rightness can be understood as conformity to deontological standard(s). The same goes for epistemology. The truth norm could be understood as a deontological standard of the same sort. Just as (a) and (b) are compatible with moral deontology, so Alston s view is compatible with a truth first epistemic deontology, which I ll describe further in the next section. The more fundamental point here is the simple one emphasized earlier that one intuitively attractive theory can be underpinned by different normative foundations. Indicator reliabilism can be underpinned by different conceptions of how truth governs belief. A non-consequentialist way to think of how truth governs belief is to think that there is an objective norm of correctness for belief it is correct to believe p only if p and that good epistemic reasons for belief are objective evidence that the belief that p would be in conformity with this norm. Of course, one might frame indicator reliabilism in a gerrymandered consequentialist way. One could take the fundamental epistemic norm not to be a deontological norm of correctness but rather a consequentialist principle that says Promote a high ratio of true to false beliefs!, and then preserve the epistemic analogues of (a) and (b). But this fact merely shows that a plausible non-consequentialist view in epistemology can be extensionally equivalent to a gerrymandered consequentialist view. The case-based evidence for indicator reliabilism doesn t decide between non-consequentialist and gerrymandered consequentialist framings. One might try to bracket indicator reliabilism on the grounds that it is not a purely reliabilist view, but rather a hybrid with internalist and externalist elements. It was intended as such a hybrid in Alston (1988). But the mere fact that a view has internalist and externalist elements is not a sufficient reason for disqualifying it as a form of reliabilism. Goldman (1986) s theory is reliabilist if any theory is. But it allows internalist factors to defeat externalist prima 19

20 facie justification. There is admittedly an important distinction between views on which reliability is merely necessary for justification and views on which it is necessary and sufficient for justification. But the former views are full-fledged externalisms if the reliability required is reliability vis-à-vis external world propositions. In any case, the most popular versions of reliabilism about justification don t hold that reliable formation is necessary and sufficient for justified belief across the board. Even Lyons (2009) perhaps the most mad-dog reliabilist around rejects this view for non-basic beliefs. So much for how indicator reliabilism could be given non-consequentialist foundations. I ll conclude with a brief mention of 80s reliabilism about knowledge. Dretske s Knowledge and the Flow of Information was as important an event for externalist epistemology as Goldman s Epistemology and Cognition. Dretske (1981: Ch.4) continued in the footsteps of the early Goldman, denying that knowledge should be analyzed in terms of justification, and instead offering a synthesis of earlier causal accounts of knowledge with insights from the indicator reliabilist tradition. This view is a further reminder that reliabilism can be understood as a non-normative theory, and hence not a consequentialist theory. 3.4 Virtue Reliabilism Another theory often classified as a version of reliabilism is the virtue epistemology of Sosa (1991, 2007, 2015) and Greco (2010); indeed, when this view was first introduced at the end of The Raft and the Pyramid, the section title used to introduce it was Reliabilism: An Ethics of Moral Virtues and an Epistemology of Intellectual Virtues. But this view is not a version of epistemic consequentialism. If this view is intended not only as an alternative to other forms of reliabilism, but as genuinely analogous to virtue ethics (hence earning its name), it is best read as having two levels, one foundational and the other factoral. At the foundational level, the account is the epistemic analogue of virtue ethics. As a result, it is analogous to an alternative to the other major foundational options, which are epistemic consequentialism and deontology. At the factoral level, the account is extensionally equivalent to a form of reliabilism in which person-level dispositions worthy of the title competences rather than potentially subpersonal processes are the 20